Image Sensor: Did you visit Image Sensors 2008 conference this year? What's your opinion on the presentation by Toshiba?
I find Toshiba proverbial tortoise in the image sensor race! Slow and steady. Slow yet definite/firm steps. Any opinions from the readers and the blogger?
Basically, Toshiba's presentation says that image sensor industry is evolving toward consolidation because this is what has happened with other similar industries.
True, image sensors are similar to other industries, but with a twist: they are moving full steam ahead into the brick wall. The pixel shrink race, tortoise or not, can not continue beyond 1.1-1.4um, not in its current form anyway. Then few things could possibly happen:
- pixel shrink stops. It means that in 3-4 years there will be no differentiator between image sensor companies and fabs/foundries. In that case I agree that there will be a big consolidation and whole the industry will go to the cheapest to manufacture and design country, be it China, Vietnam or Africa by that time.
- A new technology arises that allows the shrink path to continue. The possible candidates can be BSI, film on silicon (like e-Phocus or Fujifilm), high full well (like Advasense) or Digital Jot (by Eric Fossum), or, most probably, a combination of them. Then the move to consolidation would be delayed by, may be, 20 years.
Interesting view. You opine that there could be a disruption of some sort by a new technology that can suddenly upset the incumbent "titans" (to quote Toshiba). Right?
But, the moot question is - whether any of these disruptive technologies you have mentioned can bring down the cost per unit without driving the performance down. If the new technologies do not reduce the cost, then they will be limited to niche applications.
See in the memory business... Samsung and Toshiba are volume leaders. There are a number of innovative technologies. But, none can scale to Samsung or Toshiba's size!
Just remember that Toshiba was the first large company to sell CMOS (active pixel) image sensors and provided the first real validation for Photobit, VLSI Vision, and Omnivision. To say Toshiba is a slow tortoise is not historically accurate. -EF
Also, a new technology needs to provide a compelling advantage, one of which could be cost, but it could also be power or size or new functionality, etc., in order to displace the incumbent technology.
At the beginning it was clear to me that CMOS image sensor would also have a cost advantage, eventually, perhaps measured at the system level, but the real leverage came from power and camera on a chip aka SOC to reduce camera electronics footprint.
Hi EF (good to talk to you, you're so famous!) : problem now is not to check if CMOS sensor have a cost advantage. They have it. Now the fact is what CMOS technology is cost effective vs other CMOS technologies. About BSI, yes it is good for pixel shrink, of course, but as you may know, it is costly (SOI!) if you want to have same image quality as FSI. I really don't think OVT BSI technology offers good image quality : it is a marketing buzz for me. Besides, BSI does not appears as necessary for 1.4µm pixels. The winners for 1.1µm pixel will be the guys who masterise BSI, without SOI, and with good image quality. And I am not sure they are identified already. Regarding smaller pixel sizes (0.9µm and less), the 1T pixel approach is a serious competitor to BSI and FSI standard 4T pixels.
I don't agree with your opinions, but it is not important. I do think BSI is just about here and will stay for a while. I think you underestimate the volume vs. manufacturing cost relationship when it comes to BSI.
I am curious what 1T approach you think is a serious competitor these days to 4T? CMD and BCMD have many subtle problems. And 4T is actually less than 4T when you consider sharing RO (first invented by Eid, as I recall, at AT&T Bell Labs.)
Dear EF, I think important and constructive you do not agree with me. Especially when we share the same statement : I did not say BSI is not the near future, but a good and affordable BSI is not available for market now. A cost-effective one exists, and a good one also. Mix of them is not available for market now, even if some players are claiming it. Sharing RO is common today, but 1T means no Transfer gate, Reset, Source follower and readout, even shared, all together. It is much more ambitious and we all should think about it !
DRAMs are in better situation than image sensors. At least at each process node DRAM cell size becomes smaller. So, whoever develops a more advanced process faster, gets smaller, faster and lower power chips. Toshiba and Samsung, being among the process shrink leaders, can provide cheaper and better DRAMs.
Not so in image sensors. Even at 22nm process node I doubt that the "conventional" 4T pixel of 1.1um size can provide an acceptable level of performance, unfortunately.
So, if no disruptive technology is found, in few years all the players would come to the limit of the today's pixel and compete on price only. It's not DRAM market, it's more like cheap plastic toys one. Less probable, but also possible, image sensor would become a no-value building block of more complex chips, something like PLL. I recall in 80s having integrated PLL in a microprocessor was a big differentiator, while today one can buy it from a cheap IP provider for $4K.
So, the industry needs this disruptive technology to live, at least to live in its current form.
To EF:
Talking about the possible technologies (organic or inorganic thin films, BSI, Digital Jot, high full well), I do not see how they alone can save cost and power. However, when put in the context of general process shrink, it can bring us a very good quality integrated ISP with 1M gates, which is not feasible to integrate onto image sensor now.
To Anonymous:
With current 4-pixel sharing and SEL transistor elimination, we have 1.25 transistor per pixel. Is this a big deal to go down to 1T from here?
What I tried to say was that a new technology should bring some advantage, cost or power or size or also performance or functionality. In fact the typical product model is to increase performance and functionality in order to keep selling price constant. Usually the cost and ASP go thru some curve over the product life where gross margin hovers between 25% and 35% (for large volume products).
The magic bullet for any one company would be a great performance and/or functionality increase obtained using special process technology not generally available to other manufacturers. That allows a higher ASP and gross margin. Perhaps CIGS is one of those. BSI is a little like that. CCDs were a little like that. High speed CMOS sensors were and still are a little like that. In any case, the advantage to replace an incumbent technology has to be compelling, not just a bit better. To me, a stacked structure (e.g. BSI + X3) with digital jot RO would be in that category.
Fossum Rule #1. Never do in analog what one can do digitally in the next few years.
Fossum Rule #2. Never underestimate the force of marketing. Marketing > Engineering.
Fossum Rule #3. Never underestimate the power of continuous improvement.
"Fossum Rule #2. Never underestimate the force of marketing. Marketing > Engineering."
Management and pointy-haired bosses have a lot of power in the dark side of the Force too, but nothing like a Sith Lord of Marketing. Imagine a world where the Council of Jedi Engineers were in control...
Dear EF: Thanks for sharing your opinions. As an industry old timer and veteran, your views matter. And, I agree with your three rules. Esp. Rule #2. Great engineering without any confidence of making a promise is USELESS. Contrarily - Not-so-great engineering with the confidence to promise something is what makes sense. And, That's marketing.
There is an old saying that marketing is responsible for all successes, while engineering is responsible for all failures.
It holds pretty well in image sensors. If we look on the recent flops - Magnachip, Newport Imaging, Y-Media, CMOX, Micro One, Biomorphic - they all failed because of engineering problems (in Magnachip's case production also took part).
Hey Image Sensor: I never said don't underestimate good engineering. All I am saying is Good Engineering is a must have. Good Marketing is the confidence to say I Promise to deliver a breakthrough performance. "The confidence" separates good marketing from bad marketing.
Both engineering and marketing play a complementary role. Good marketing with poor engineering is vaporware. Good engineering with poor marketing is just a good acquisition target (at best).
Image Sensor: Did you visit Image Sensors 2008 conference this year? What's your opinion on the presentation by Toshiba?
ReplyDeleteI find Toshiba proverbial tortoise in the image sensor race! Slow and steady. Slow yet definite/firm steps. Any opinions from the readers and the blogger?
Basically, Toshiba's presentation says that image sensor industry is evolving toward consolidation because this is what has happened with other similar industries.
ReplyDeleteTrue, image sensors are similar to other industries, but with a twist: they are moving full steam ahead into the brick wall. The pixel shrink race, tortoise or not, can not continue beyond 1.1-1.4um, not in its current form anyway. Then few things could possibly happen:
- pixel shrink stops. It means that in 3-4 years there will be no differentiator between image sensor companies and fabs/foundries. In that case I agree that there will be a big consolidation and whole the industry will go to the cheapest to manufacture and design country, be it China, Vietnam or Africa by that time.
- A new technology arises that allows the shrink path to continue. The possible candidates can be BSI, film on silicon (like e-Phocus or Fujifilm), high full well (like Advasense) or Digital Jot (by Eric Fossum), or, most probably, a combination of them. Then the move to consolidation would be delayed by, may be, 20 years.
Interesting view. You opine that there could be a disruption of some sort by a new technology that can suddenly upset the incumbent "titans" (to quote Toshiba). Right?
ReplyDeleteBut, the moot question is - whether any of these disruptive technologies you have mentioned can bring down the cost per unit without driving the performance down. If the new technologies do not reduce the cost, then they will be limited to niche applications.
See in the memory business... Samsung and Toshiba are volume leaders. There are a number of innovative technologies. But, none can scale to Samsung or Toshiba's size!
Just remember that Toshiba was the first large company to sell CMOS (active pixel) image sensors and provided the first real validation for Photobit, VLSI Vision, and Omnivision.
ReplyDeleteTo say Toshiba is a slow tortoise is not historically accurate.
-EF
Also, a new technology needs to provide a compelling advantage, one of which could be cost, but it could also be power or size or new functionality, etc., in order to displace the incumbent technology.
ReplyDeleteAt the beginning it was clear to me that CMOS image sensor would also have a cost advantage, eventually, perhaps measured at the system level, but the real leverage came from power and camera on a chip aka SOC to reduce camera electronics footprint.
I feel like the old guy now.
-EF
Hi EF (good to talk to you, you're so famous!) : problem now is not to check if CMOS sensor have a cost advantage. They have it. Now the fact is what CMOS technology is cost effective vs other CMOS technologies.
ReplyDeleteAbout BSI, yes it is good for pixel shrink, of course, but as you may know, it is costly (SOI!) if you want to have same image quality as FSI. I really don't think OVT BSI technology offers good image quality : it is a marketing buzz for me. Besides, BSI does not appears as necessary for 1.4µm pixels.
The winners for 1.1µm pixel will be the guys who masterise BSI, without SOI, and with good image quality.
And I am not sure they are identified already.
Regarding smaller pixel sizes (0.9µm and less), the 1T pixel approach is a serious competitor to BSI and FSI standard 4T pixels.
I don't agree with your opinions, but it is not important. I do think BSI is just about here and will stay for a while. I think you underestimate the volume vs. manufacturing cost relationship when it comes to BSI.
ReplyDeleteI am curious what 1T approach you think is a serious competitor these days to 4T? CMD and BCMD have many subtle problems. And 4T is actually less than 4T when you consider sharing RO (first invented by Eid, as I recall, at AT&T Bell Labs.)
-EF
Dear EF, I think important and constructive you do not agree with me. Especially when we share the same statement : I did not say BSI is not the near future, but a good and affordable BSI is not available for market now. A cost-effective one exists, and a good one also. Mix of them is not available for market now, even if some players are claiming it.
ReplyDeleteSharing RO is common today, but 1T means no Transfer gate, Reset, Source follower and readout, even shared, all together. It is much more ambitious and we all should think about it !
To Toshiba Analyst:
ReplyDeleteDRAMs are in better situation than image sensors. At least at each process node DRAM cell size becomes smaller. So, whoever develops a more advanced process faster, gets smaller, faster and lower power chips. Toshiba and Samsung, being among the process shrink leaders, can provide cheaper and better DRAMs.
Not so in image sensors. Even at 22nm process node I doubt that the "conventional" 4T pixel of 1.1um size can provide an acceptable level of performance, unfortunately.
So, if no disruptive technology is found, in few years all the players would come to the limit of the today's pixel and compete on price only. It's not DRAM market, it's more like cheap plastic toys one. Less probable, but also possible, image sensor would become a no-value building block of more complex chips, something like PLL. I recall in 80s having integrated PLL in a microprocessor was a big differentiator, while today one can buy it from a cheap IP provider for $4K.
So, the industry needs this disruptive technology to live, at least to live in its current form.
To EF:
Talking about the possible technologies (organic or inorganic thin films, BSI, Digital Jot, high full well), I do not see how they alone can save cost and power. However, when put in the context of general process shrink, it can bring us a very good quality integrated ISP with 1M gates, which is not feasible to integrate onto image sensor now.
To Anonymous:
With current 4-pixel sharing and SEL transistor elimination, we have 1.25 transistor per pixel. Is this a big deal to go down to 1T from here?
What I tried to say was that a new technology should bring some advantage, cost or power or size or also performance or functionality. In fact the typical product model is to increase performance and functionality in order to keep selling price constant. Usually the cost and ASP go thru some curve over the product life where gross margin hovers between 25% and 35% (for large volume products).
ReplyDeleteThe magic bullet for any one company would be a great performance and/or functionality increase obtained using special process technology not generally available to other manufacturers. That allows a higher ASP and gross margin. Perhaps CIGS is one of those. BSI is a little like that. CCDs were a little like that. High speed CMOS sensors were and still are a little like that.
In any case, the advantage to replace an incumbent technology has to be compelling, not just a bit better.
To me, a stacked structure (e.g. BSI + X3) with digital jot RO would be in that category.
Fossum Rule #1. Never do in analog what one can do digitally in the next few years.
Fossum Rule #2. Never underestimate the force of marketing. Marketing > Engineering.
Fossum Rule #3. Never underestimate the power of continuous improvement.
"Fossum Rule #2. Never underestimate the force of marketing. Marketing > Engineering."
ReplyDeleteManagement and pointy-haired bosses have a lot of power in the dark side of the Force too, but nothing like a Sith Lord of Marketing. Imagine a world where the Council of Jedi Engineers were in control...
To Eric Fossum:
ReplyDeleteWell said, I agree with everything.
Dear EF: Thanks for sharing your opinions. As an industry old timer and veteran, your views matter. And, I agree with your three rules. Esp. Rule #2. Great engineering without any confidence of making a promise is USELESS. Contrarily - Not-so-great engineering with the confidence to promise something is what makes sense. And, That's marketing.
ReplyDeleteThis comment has been removed by the author.
ReplyDeleteTo Toshiba analyst:
ReplyDeleteThere is an old saying that marketing is responsible for all successes, while engineering is responsible for all failures.
It holds pretty well in image sensors. If we look on the recent flops - Magnachip, Newport Imaging, Y-Media, CMOX, Micro One, Biomorphic - they all failed because of engineering problems (in Magnachip's case production also took part).
So... never underestimate a good engineering.
Hey Image Sensor: I never said don't underestimate good engineering. All I am saying is Good Engineering is a must have. Good Marketing is the confidence to say I Promise to deliver a breakthrough performance. "The confidence" separates good marketing from bad marketing.
ReplyDeleteBoth engineering and marketing play a complementary role. Good marketing with poor engineering is vaporware. Good engineering with poor marketing is just a good acquisition target (at best).
Great views.